Most proteins conduct their functions through interactions with other proteins. The atomic-level quaternary structure of protein-protein complexes can provide a clear physical landscape to help our understanding of how the interactions are conducted in living cells and how new therapies can be designed to regulate the interaction networks. Since experimental characterization of complex structures is difficult and expensive, computational modeling of the protein-protein interactions has been a major theme in computational biology. Most efforts have been focused on rigid-body docking, which builds complex conformations by combining known structures of interacting components. But docking is applicable only when the monomer structures are known and the success rate is low when components involve conformational change upon binding. Alternatively, complex structures can be deduced from homologous structures with alignments generated by the multi-chain threading technique. While the latter approach has the advantage of not requiring solved monomer structures, the modeling accuracy for distant-homology targets is unreliable and the threading alignments generally have gaps and errors. In this project, we seek to develop a new generation of computational approaches aiming to significantly improve the coverage and accuracy of protein-protein complex structure modeling by the integration of the cutting-edge rigid-body docking and threading assembly simulations. The specific aims include: (1) development of new interface-specific threading algorithms for distant-homology detection; (2) new fragment assembly simulation method for full-length complex structure construction and refinement; (3) development of new strategies for ab initio docking; (4) integration of the threading and docking methods for low-resolution docking and template-based docking structure refinement. The algorithms will be systematically trained on large-scale benchmark protein sets and tested in community-wide docking experiments, with focus on modeling the binding-induced conformational changes and predicting high-resolution complex structures for distantly homologous proteins. The methods and potentials developed in this project will be made freely available to the general community through Internet websites. The long-term goals of this project are (a) to develop advanced computer methods for accurate structure modeling of various protein-protein complexes, and (b) to utilize the methods for genome-wide structure modeling and structure-based function annotation of protein-protein networks of various organisms.